Nissim and Penman - Ratio Analysis and Equity Valuation From Research To Practice

Review of Accounting Studies, 6, 109154, 2001
C 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.

Ratio Analysis and Equity Valuation:
From Research to Practice
DORON NISSIM dn75@columbia.edu
Columbia University, Graduate School of Business, 3022 Broadway, Uris Hall 604, New York, NY 10027
STEPHEN H. PENMAN shp38@columbia.edu
Columbia University, Graduate School of Business, 3022 Broadway, Uris Hall 612, New York, NY 10027
Abstract. Financial statement analysis has traditionally been seen as part of the fundamental analysis required
for equity valuation. But the analysis has typically been ad hoc. Drawing on recent research on accounting-based
valuation, this paper outlines a nancial statement analysis for use in equity valuation. Standard protability
analysis is incorporated, and extended, and is complemented with an analysis of growth. An analysis of operating
activities is distinguished from the analysis of nancing activities. The perspective is one of forecasting payoffs
to equities. So nancial statement analysis is presented as a matter of pro forma analysis of the future, with
forecasted ratios viewed as building blocks of forecasts of payoffs. The analysis of current nancial statements is
then seen as a matter of identifying current ratios as predictors of the future ratios that determine equity payoffs.
The nancial statement analysis is hierarchical, with ratios lower in the ordering identied as ner information
about those higher up. To provide historical benchmarks for forecasting, typical values for ratios are documented
for the period 19631999, along with their cross-sectional variation and correlation. And, again with a view to
forecasting, the time series behavior of many of the ratios is also described and their typical long-run, steady-state
levels are documented.
Keywords: Financial statement analysis, ratio analysis, equity valuation
It goes almost without saying that, in an applied discipline like accounting, the aim of
research is to affect practice. Theory can be admired on a number of dimensions, but a stream
of research is ultimately judged on the products it delivers, how it enhances technology.
Engineering and medical research, to name just two endeavors, have this orientation. Our
colleagues in nance have been successful in product development. While making major
contributions to economic theory, they have also engineered such products as derivative
pricing, risk measurement, hedging instruments, portfolio insurance, and asset allocation,
some, to be sure, more successful than others.
Inthe area of equityanalysis, researchinnance has not beensuccessful. Equityanalysis
or fundamental analysiswas once the mainstream of nance. But, while enormous steps
have been taken in pricing derivatives on the equity, techniques to value equities have not
advanced much beyond applying the dividend discount model. So-called asset pricing mod-
els, like the Capital Asset Pricing Model, have been developed, but these are models of risk
and the expected return, not models that instruct how to value equities. Real option anal-
ysis has been applied to equity valuation, but the measurement problems are signicant.
Some progress has been made by accounting researchers in what has come to be referred
to as accounting-based valuation research. That is not surprising. Equity analysis is largely
an analysis of information, and accountants deal with information about rms. This paper
carries the recent research to the level of product design.
110 NISSIM AND PENMAN
Traditional fundamental analysis (before modern nance) was very much grounded in the
nancial statements. So, for example, Graham, Dodd and Cottles Security Analysis (1962)
is not the security analysis of modern nance texts (that involves the analysis of prices, beta
estimation, and asset allocation), but rather security analysis that analyzes fundamentals
through the nancial statements. However, nancial statement measures were linked to
equity value in an ad hoc way, so little guidance was given for understanding the implications
of a particular ratioa prot marginor aninventoryturnover, for examplefor equityvalue.
Nor was a comprehensive scheme advanced for identifying, analyzing and summarizing
nancial statement information in order to draw a conclusion as to what the statements, as
a whole, say about equity value. Equity value is determined by future earnings power,
it was said, but there was no explicit justication for using future earnings as a valuation
attribute, nor was there explicit development of the forecasting of this earnings power.
A considerable amount of accounting research in the years since Graham, Dodd and
Cottle has been involved in discovering hownancial statements informabout equity value.
The whole endeavor of capital markets research deals with the information content of
nancial statements for determining stock prices. The extensive time-series-of-earnings
literature summarized in Brown (1993) focuses on forecasting earnings, often with valuation
inmind. Papers suchas Lipe (1986), Ou(1990), OuandPenman(1989), LevandThiagarajan
(1993) and Faireld, Sweeney and Yohn (1996), to name just a few, have examined the role
of particular nancial statement components and ratios in forecasting. But it is fair to say
that the research has been conducted without much structure. Nor has it produced many
innovations for practice. Interesting, robust empirical correlations have been documented,
but the research has not produced a convincing nancial statement analysis for equity
valuation. Indeed the standard textbook schemes for analyzing statements, such as the
DuPont scheme, rarely appear in the research.
Drawing on recent research on accounting-based valuation, this paper ventures to produce
a structural approach to nancial statement analysis for equity valuation. The structure not
only identies relevant ratios, but also provides a way of organizing the analysis task. The
result is a fundamental analysis that is very much grounded in the nancial statements;
indeed, fundamental analysis is cast as a matter of appropriate nancial statement analysis.
The structural approach contrasts to the purely empirical approach in Ou and Penman
(1989). That paper identied ratios that predicted earnings changes in the data; no thought
was given to the identication. The approach also contrasts to that in Lev and Thiagarajan
(1993) who defer to expert judgment and identify ratios that analysts actually use in
practice.
Valuation involves forecasting payoffs. Forecasting is guided by an equity valuation model
that species what is to be forecasted. So, for example, the dividend discount model directs
the analyst to forecast dividends. Because it focuses on accrual-accounting nancial state-
ments, the residual income valuation model, recently revived through the work of Ohlson
(1995) and Feltham and Ohlson (1995), serves as an analytical device to organize thinking
about forecasting and analyzing nancial statements for forecasting. This model is a state-
ment of how book value and forecasted earnings relate to forecasted dividends and thus
to value. The ratio analysis in this paper follows from recognition of standard accounting
relations that determine how components of the nancial statements relate to earnings and
book values.
RATIO ANALYSIS AND EQUITY VALUATION 111
Our focus on the residual income valuation model is not to suggest that this model is the
only model, or even the best model, to value equities. Penman (1997) shows that dividend
and cash-ow approaches give the same valuation as the residual income approach under
certain conditions. The residual income model, based as it is on accrual accounting, is
of particular help in developing an analysis of accrual-accounting nancial statements.
But cash ows and dividends are tied to accrual numbers by straightforward accounting
relations, so building forecasts of accrual accounting numbers with the aid of analysis builds
forecasts of free cash ows and dividends also, as will be seen.
The scheme is not offered as the denitive way of going about nancial statement analysis.
There is some judgment as to what makes sense. This is inevitably part of the art of design
in bringing academic models to practice (Colander (1992)). As such it stands as a point of
departure for those with better judgment.
The paper comes in two parts. First it identies ratios that are useful for valuation. Second,
it documents typical values of the ratios during the period 1963 to 1999.
Identication. Residual earnings valuation techniques are so-called because equity value
is determined by forecasting residual income. As a matter of rst order, ratio identication
amounts to identifying ratios that determineor drivefuture residual income so that, by
forecasting these ratios, the analyst builds a forecast of residual income. So relevant ratios
are identied as the building blocks of a forecast, that is, as the attributes to be forecasted
in order to build up a forecast of residual income. However, ratios are usually seen as
information in current nancial statements that forecasts the future. So current nancial
statement ratios are deemed relevant for valuation if they predict future ratios. Accordingly
the identication of (future) residual income drivers is overlaid here with a distinction
between transitory features of ratios (that bear only on the present) and permanent
features (that forecast the future).
At the core is an analysis of protability. Many of the standard protability ratios are
included, so many aspects of the analyses are familiar. Indeed the paper serves to inte-
grate protability analysis with valuation. But rened measures of operational protability
are presented and an alternative analysis of leverage is introduced. And protability ratios
are complemented with ratios that analyze growth, for both protability and growth drive
residual earnings.
Not only are relevant ratios identied, but an algebralike the traditional DuPont analysis
(which is incorporated here)ties the ratios together in a structured way. This algebra not
only explains howratios sumup as building blocks of residual income but also establishes
a hierarchy so that many ratios are identied as ner information about others. So the analyst
identies certain ratios as primary and considers other ratios down the hierarchy only if
they provide further information. This brings an element of parsimony to practical analysis.
But it also provides a structure to researchers who wish to build (parsimonious) forecasting
models and accounting-based valuation models.
In residual income valuation, forecasted income must be comprehensive income, other-
wise value is omitted. So the ratio analysis is based on a comprehensive income statement.
This is timely because FASB Statement No. 130 now requires the reporting of compre-
hensive income on a more transparent basis, and other recent FASB statements, notably
statements 115 and 133, have introduced new components of comprehensive income. But
comprehensive income contains both permanent and transitory components of income. For
forecast and valuation, these need to be distinguished.
The analysis makes a separation between operating and nancing items in the nancial
statements. This is inspired by the Modigliani and Miller notion that it is the operating
activities that generate value, and that apart frompossible tax effects, the nancing activities
are zero net-present-value activities. The separation also arises from an appreciation that
nancial assets and liabilities are typically close to market value in the balance sheet (more
so since FASB Statement No. 115) and thus are already valued. But not so for the operating
assets and liabilities, so it is the operating activities that need to be analyzed. The distinction
is a feature of the accounting-based valuation model in Feltham and Ohlson (1995) and of
economic prot versions of the residual income model.
The distinction between operating and nancing activities requires a careful separation of
operating and nancing items in the nancial statement that leads, in the paper, to a rened
measure of operating protability to the one often advanced in texts. It also leads to better
understanding of balance-sheet leverage that involves two leverage measures, one arising
from nancing activities, the other from operating activities. And it isolates growth as an
attribute of the operating activities, not the nancing activities, and develops measures of
growth from the analysis of the operating activities.
Documentation. Ratio analysis usually compares ratios for individual rms against bench-
marks from comparable rmsboth in the past and the presentto get a sense of what is
normal and what is abnormal. The paper provides a historical analysis of ratios that
yields such benchmarks for the equity researcher using residual earnings techniques.
Appreciating what is typical is of assistance in developing prior beliefs for any forecasting,
and particularly so in a valuation context because there is a tendency for many of the relevant
ratios to revert to typical values over time, as will be seen. Further, valuation methods that
involve forecasting require continuing value calculations at the end of a forecast period.
These calculations require an assessment of a steady state for residual income and are
often seen as problematical. The documentation here gives a sense of the typical steady state
for the drivers of residual income and thus a sense of the typical terminal value calculations
required. It shows that steady-state conditions typically occur within reasonable forecast
horizons and their form is similar to that prescribed by residual income models. This gives
a level of comfort to those applying residual income techniques.
The documentation also helps in the classication of nancial statement items into
permanent and transitory. This classication inevitably involves some judgment but
the displays here give typical fade rates for the components of residual income drivers
and thus an indication of which components are typically transitory.
1. The Residual Earnings Valuation Model
There are many ratios that can be calculated from the nancial statements and the equity
analyst has to identify those that are important. The residual earnings equity valuation model
brings focus to the task.
The residual earnings model restates the non-controversial dividend discount model. Rec-
ognizing the (clean surplus) relation that net dividends are always equal to comprehensive
income minus the change in the book value of equity, the model (with no further assump-
tion) expresses value in terms of accounting numbers rather than forecasted dividends. The
model can be applied to the valuation of any asset but the focus here is on the common
equity. The model states the value of common equity at date 0 as
V
E
0
= CSE
0
+
t =1
t
E
(CNI
t
(
E
1) CSE
t 1
) (1)
where CSE is the book value of common equity, CNI is comprehensive (net) income
available to common,
E
is one plus the required return for common equity (the eq-
uity cost of capital), and t is a counter of future years beyond the current year, year 0.
CNI
t
(
E
1) CSE
t 1
is residual earnings or residual income and we will refer to it as
RE
t
. Bars over numbers indicate they are forecasted amounts. The determination of the cost
of capital, though important to the valuation exercise, is not addressed here; we represent it
simply as an unknown constant.
The innite-horizon forecasting required by this model is considered impractical. So, in
actual analysis, forecasts are made for a nite number of years and a continuing value,
CV
T
, is added at the forecast horizon, T:
V
E
0
= CSE
0
+
T
t =1
t
E
RE
t
+
CV
T
T
E
. (1a)
As the continuing value is the value at T of residual earnings beyond T, it is equal to
V
E
T
CSE
T
, that is, the forecasted premium at T. Continuing values typically take three
forms:
CV
T
= 0 (CV1)
CV
T
= RE
T+1
/(
E
1) (CV2)
CV
T
= RE
T+1
/(
E
g) (CV3)
where g is one plus the rate of growth in expected residual earnings. CV1 forecasts a steady
state of zero residual earnings after T; CV2 forecasts non-zero but constant steady-state
residual earnings after T; and CV3 forecasts perpetual growth in expected residual earnings
after T. CV2 is the no-growth case of CV3.
This model is well established in the academic literature (in Preinreich (1938), Kay (1976),
Edwards and Bell (1961) and Ohlson (1995), for example). It has been applied in recent
valuation research and nancial statement analysis (in Brief and Lawson (1992), Frankel
and Lee (1998), Lee, Myers and Swaminathan (1999), Penman and Sougiannis (1998),
Francis, Olsson and Oswald (2000), and Abarbanell and Bernard (2000), for example), and
has increasing currency in nancial analysis texts and practical equity research.
To apply the model the analyst must develop forecasts of RE. He or she must also decide
which version of the continuing value is appropriate and at what point in the future it is to be
applied. Forecasting the long-term growth rate is of particular importance. The nancial
statement analysis in the next section gives a framework for developing RE forecasts, a way
of breaking down RE into components to be forecasted, and an orderly way of assembling
information to forecast these components. The documentation in Section 3 gives historical
values for the components and evidence on steady state for RE and its components.
2. Ratio Identication
Residual earnings compares earnings to net assets employed and so is a measure of prof-
itability. Residual earnings can be expressed in ratio form as:
RE
t
= [ROCE
t
(
E
1)]CSE
t 1
where ROCE
t
= CNI
t
/CSE
t 1
is the rate of return on common equity. So forecasting
residual earnings involves forecasting ROCE and book values to be put in place to earn
the forecasted ROCE. Distinguishing ROCE and book value as two separate attributes to
forecast helps to compartmentalize the task. But this is not to mean that return on book
values and book values are independent. Formally, while this expression holds for realized
returns, it is not the case that
E[CNI
t
(
E
1)CSE
t 1
] =
CNI
t
CSE
t 1
(
E
1)
E(CSE
t 1
),
unless CNI
t
/CSE
t 1
and CSE
t 1
are uncorrelated. The amount of equity investment might
depend on ROCE and the accounting for book values may affect ROCE. Under conservative
accounting, for example, ROCE is below its no-growth rate if investments are growing, and
reducing investments generates higher ROCE, as modeled in Beaver and Ryan (2000) and
Zhang (2000). Strictly, the forecast is of expected book values and expected earnings on
expected book values. Accordingly, forecasting is done as a matter of scenario analysis:
ROCE and book values are forecasted for alternative scenarios, producing forecasted CNI
and CSE for each scenario, then averages are taken over probability-weighted scenarios.
So the analysis here should be seen as one for developing forecast scenarios.
1
Contingent
scenarios can be incorporated so that scenarios involving real options, growth options and
adaptation options are thus accommodated by the analysis.
2.1. The Drivers of Return on Common Equity (ROCE)
ROCE is the summary protability ratio in nancial statements and is driven by in-
come statement line items that sum to net income in the numerator and balance sheet
items that sum to the net assets in the denominator. Residual income valuation requires
that forecasted income be comprehensive income, otherwise value is lost. So our income
statement analysis is of all the line items that sum to comprehensive income. Our analy-
sis also distinguishes operating protability from the protability identied with the -
nancing activities. As is standard, operating activities are those involved in producing
goods and services for customers. Financing activities have to do with raising cash for
the operations and disposing of cash from the operations.
2
A division of line items that
distinguishes operating from nancial activities is a starting point for analysis of ROCE
drivers:
CNI = Comprehensive Operating Income (OI)
Comprehensive Net Financial Expense (NFE) (2)
CSE = Net Operating Assets (NOA) Net Financial Obligations (NFO) (3)
where
NFE = (Financial Expense Financial Income), after tax
NOA = Operating Assets (OA) Operating Liabilities (OL)
NFO = Financial Obligations (FO) Financial Assets (FA).
Operating liabilities are those generated by operations (like accounts payable, wages pay-
able, pension liabilities and deferred tax liabilities), while nancial liabilities are those from
raising funds to nance operations. Financial assets (bonds held) are available to nance
operations and effectively reduce debt to nance operations (bonds issued). Balance sheet
totals are maintained; that is,
Total Assets = OA +FA,
Total Liabilities & Preferred Stock = OL +FO,
so all balance sheet items are assigned to a category.
Net nancial expense (NFE) is the (comprehensive) net expense owing fromnet nancial
obligations and includes interest expense minus interest income, preferred dividends, and
realizedandunrealizedgains andlosses onnancial assets andobligations; all items drawing
tax or tax benets are multiplied by (1 marginal tax rate) unless reported on an after-tax
basis. All accounting items are identied from the common shareholders point of view.
Thus preferred dividends are a nancial expense and preferred stock is a nancial obligation.
If a rm has net nancial assets rather than net nancial obligations (nancial assets are
greater than nancial obligations) then it generates net nancial income rather than net
nancial expense.
Operating income (OI = CNI + NFE) is the income owing from net operating assets
and, by the calculations here, is after tax.
Comparing each income statement component to its corresponding balance sheet com-
ponent yields measures of operating protability and nancing protability:
Return on Net Operating Assets (RNOA)
t
=
OI
t
NOA
t 1
and
Net Borrowing Cost (NBC)
t
=
NFE
t
NFO
t 1
.
RNOAis different fromthe more commonreturnonassets. Returnonassets include nancial
assets in its base and excludes operating liabilities, so it confuses operating and nancing
activities. RNOA is similar to Return on Invested Capital (ROIC) but we use RNOA here
to emphasize that operating liabilities reduce the net operating assets employed. If the rm
is a net creditor rather than a net debtor (nancial assets greater than nancial obligations,
so NFO is negative), NBC is return on net nancial assets.
Given the accounting equation (3) and relation (2),
ROCE =
NOA
CSE
RNOA
NFO
CSE
NBC
,
that is, ROCE is a weighted average of the return on operating activities and the return on
nancing activities. It is understood, unless otherwise indicated, that all measures are for the
same period. Income statement amounts are for the period and balance sheet amounts are
for the beginning of the period. (Balance sheet amounts can also be averages for a period,
as is common practice.) Rearranging terms,
ROCE = RNOA + [FLEV SPREAD] (4)
where
FLEV =
NFO
CSE
(Financial Leverage)
and
SPREAD = RNOA NBC.
Thus ROCE is driven by the return on operations with an additional return from the
leverage of nancial activities. This leverage effect is determined by the amount of lever-
age and the spread between the return on operations and net borrowing costs. A further
decomposition yields
ROCE = [PM ATO] + [FLEV SPREAD] (4a)
where
PM= OI/Sales (Prot Margin)
and
ATO = Sales/NOA. (Asset Turnover)
This decomposition of RNOA into PM and ATO follows the standard DuPont analysis.
Note that the asset turnover is different from the more common measure of sales/assets.
Following that standard analysis further, PM can be broken down into the gross margin
ratio and expense/sales ratios, and ATO into turnover ratios for individual operating assets
and liabilities.
There are some modications, however. Operating income includes income generated
from sales, after expenses, and thus PM captures the protability of each dollar of sales.
But it also includes items not incurred to generate the reported saleslike equity share
of income in a subsidiary, dividends, and gains and losses on equity investments marked
to market. We refer to these items as Other Items and exclude them from a revised prot
margin:
Sales PM= OI from Sales/Sales.
So,
ROCE = [Sales PM ATO] +
Other Items
NOA
+[FLEV SPREAD]. (5)
Both Sales PM and Other Items are after tax. Other Items/NOA has little meaning, but
protability of sales is identied without noise. Prot margins are typically regarded as
crucial and this revised prot margin cannot be affected by acquisitions accounted for under
the equity method (for example).
A further modication is required when there are minority interests in subsidiaries, for
minority interests share with the common shareholders in earnings. With minority interests
(MI) on the consolidated balance sheet, equation (3) is restated to
CSE = NOA NFO MI.
And return on total common equity is calculated as
ROTCE = (CNI + MI share of income)/(CSE + MI).
The components in (5), with FLEV redened as NFO/(CSE+MI), aggregate to ROTCE
rather than ROCE and
ROCE = ROTCE MSR (6)
where
Minority Sharing Ratio (MSR) =
CNI/(CNI + MI share of income)
CSE/(CSE + MI on balance sheet)
.
An additional driver of RNOA involves operating liabilities. Clearly the netting out of
operating liabilities in the calculation of NOA increases RNOA through a denominator
effect, and appropriately so: to the extent that a rmhas non-interest credit frompayables
(for example) it levers up its RNOA. This leverage is a driver of protability that is distinct
from nancial leverage, for it arises in the operations, not the nancial activities.
3
This
leverage can be analyzed. Suppliers who advance the payables reduce the net investment
required to run the operations and so lever up the operating protability, but suppliers
presumably charge implicitly for the credit in terms of higher prices. Denote io as the
implicit interest charge on operating liabilities other than undiscounted deferred taxes, and
calculate
ROOA =
(OI +io)
OA
as the Return on Operating Assets that would be made without leverage from operating
liabilities. Then
RNOA =
ROOA
OA
NOA
io
OL

OL
NOA
= ROOA +[OLLEV OLSPREAD]

(7)
where
OLLEV =
OL
NOA
is operating liability leverage, and
OLSPREAD = ROOA
io
OL
is operating liability spread. This is of the same form as the nancial leverage formula in
(4): RNOA is levered up by operating liability leverage and the leverage effect is deter-
mined by the operating liability leverage and the spread between ROOA and the implicit
borrowing cost. The implicit borrowing cost can be estimated with the short-termborrowing
rate.
4
Like nancial leverage, the analysis shows that operating liability leverage can be
favorable or unfavorable; the leverage is favorable only if ROOA is greater than the implicit
borrowing cost.
This analysis yields seven drivers of ROCE:
Sales Prot Margin (Sales PM)
Asset Turnover (ATO)
Other Items/NOA
Financial Leverage (FLEV)
Net Borrowing Cost (NBC) which, when compared to RNOA, gives SPREAD
Operating Liability Leverage (OLLEV)
Minority Interest Sharing (MSR)

Forecasting ROCE involves forecasting these drivers and aggregating them according to
(5) (with ROTCE substituted for ROCE), (6) and (7).
2.2. The Drivers of Book Value
To forecast residual income one must forecast CSE as well as ROCE. CSE can be decom-
posed into
CSE = Sales
NOA
Sales

CSE
NOA
and thus, when there are no minority interests,
CSE = Sales
1
ATO

1
1 +FLEV
. (8)
Sales drive the net operating assets and 1/ATO is the amount of NOA that has to be put in
place to generate a dollar of sales. The NOA can be nanced by equity or borrowing and
1/(1 +FLEV) captures this nancing decision. Accordingly, future CSE is forecasted by
predicting three drivers:
Sales
Financial Leverage (FLEV).

5
With the forecast of the drivers in (5), (6), (7) and (8), the forecasting of residual earnings
is complete. The nesting of ratios within (5), (6), (7) and (8)so that they aggregate
is by careful denition and accounting relations (2) and (3), and involve no economic
assumptions. The relationships hold under all economic conditions and for all accounting
principles provided earnings are comprehensive earnings.
2.3. Reducing the Analysis
Just as residual earnings can be calculated for common equity (net assets) so it can be cal-
culated for any component of net assets. For the two (operating and nancing) components
identied above,
Residual Operating Income (ReOI)
t
= OI
t
(
w
1) NOA
t 1
and
Residual Net Financial Expense (ReNFE)
t
= NFE
t
(
D
1) NFO
t 1
w
is the required return for the operations (we use w to donate it as the weighted-average
cost of capital, as is standard) and
D
the required return on the net nancial obligations
(the cost of capital for debt). The value of the net operating asset component of equity is
V
NOA
0
= NOA
0
+
t =1
t
w
ReOI
t
.
This is often referred to as the value of the rm or enterprise value. The value of the net
nancial obligations is
V
NFO
0
= NFO
0
+
t =1
t
D
ReNFE
t
.
By the accounting equation (applied to values rather than book values),
V
E
0
= V
NOA
0
V
NFO
0
and so V
E
0
as stated in (1) is equivalent to
6
V
E
0
= NOA
0
NFO
0
+
t =1
t
W
ReOI
t

t =1
t
D
ReNFE
t
.
For any asset or obligation measured at market value, forecasted residual income must
be equal to zero (it is forecasted to earn at the cost of capital). If NFO is measured on the
balance sheet at market value such that NFO = V
NFO
0
, then the present value of forecasted
ReNFE
t
is zero (and if the present value of ReNFE is zero, then NFO = V
NOA
0
). If so, then
for nite forecast horizons,
V
E
0
= CSE
0
+
T
t =1
t
W
ReOI
t
+
CV
NOA
T
T
W
(9)
with three forms for the continuing value:
CV
NOA
T
= 0 (CV01)
CV
NOA
T
=
ReOI
T+1
W
1
(CV02)
CV
NOA
T
=
ReOI
T+1
W
g
(CV03)
where g is now one plus the forecasted growth rate in ReOI beyond T.
Residual operating income, like residual income, can be expressed in terms of a ratio
measure of protability:
ReOI
t
= [RNOA
t
(
W
1)] NOA
t 1
so ReOI is forecasted by predicting RNOA and the net operating assets to be put in place to
earn at the forecasted RNOA. Again, operating income and NOA may not be independent
so the analysis applies to scenarios. Accordingly, with a valuation based on forecasting
ReOI, the number of drivers to be predicted are reduced. The drivers of RNOA are
Sales Prot Margin (Sales PM)
Other Items/NOA
Operating Liability Leverage (OLLEV)

As NOA = Sales/ATO, the drivers of NOA are
Sales

This reduction in the forecasting task is a clear efciency. Financial leverage considera-
tions drop out, including leverage effects on the cost of capital. One needs only forecast the
operating protability and the NOA and forget forecasting borrowing and dividend policy.
But the approach is only appropriate if nancial assets and nancial obligations are mea-
sured at their fair value. This is the case for many nancial assets under FASB Statement
No. 115. Other nancial assets and liabilities are often close to market value (as a work-
able approximation), or their fair values are disclosed in footnotes under FASB Statement
No. 107.
RNOA and NOA have a common driver (ATO). ReOI can be determined in one step that
utilizes both RNOA and NOA drivers:
ReOI = Sales
PM

W
1
ATO
(10)
The ratio, (
W
1)/ATO compares the ATO to the required return on operations and is
a measure of the efciency of the NOA in generating sales relative to the required return
on NOA.
2.4. The Drivers of Growth in Residual Earnings
Growth is an important aspect of valuation, particularly for the calculation of continuing
values. Withresidual income valuation, the appropriate growthconcept is growthinexpected
residual earnings.
7
Since residual income, for a given scenario, is driven by the accounting
rate of return and book value, its growth is driven by increases in the rate of return and/or
increases in book value. There are simplied calculations to capture this. We deal with
residual operating income, with the calculation for full residual income implicit. The growth
rate is
Growth Rate in ReOI
t
= ReOI
t
/ReOI
t 1
1 =
[RNOA
t

W
+1]NOA
t 1
[RNOA
t 1
W
+1]NOA
t 2
1.
So growth in ReOI (which can be greater or less than 0) is driven by changes in RNOA,
NOA, and the cost of capital. We deal with cases where the forecasted cost of capital is
constant.
Constant RNOA
Pro forma analysis often comes to the conclusion that, in the long-run, protability will
converge to a permanent level. If this is such that, at a point, T, in the future, one forecasts
that subsequent RNOA will equal the cost of capital, then expected ReOI is zero. But if one
forecasts that RNOA will be different from the cost of capital but constant, then
Growth Rate in ReOI
t
=
NOA
t 1
NOA
t 2
1
for all t > T. Growth is driven solely by increases in NOA.
A permanent level of non-zero ReOI (RNOA = cost of capital) may reect permanent
abnormal real protability(inthe sense that the rmcaninvest always innon-zeronet present
value projects). But it can also be induced by the accounting: conservative accounting, for
example, that always keeps book values low (by expensing R&D, for example) will yield
permanent positive ReOI even with zero NPV projects. This is modeled in Feltham and
Ohlson (1995). Beaver and Ryan (2000) provide some empirical evidence.
Constant RNOA, Constant PM and Constant ATO
As RNOA is driven by prot margin and turnover, a constant RNOA can be driven by
a changing PM with a canceling effect of changing ATO, and by both constant PM and
constant ATO. In the latter case growth in ReOI is driven solely by growth in sales. From
(10), growth in ReOI can also be expressed as
Growth Rate in ReOI
t
=
Sales
t

PM
t

W
1
ATO
t
Sales
t 1
PM
t 1

W
1
ATO
t 1
1
thus, for constant PM and ATO (and cost of capital),
Growth Rate in ReOI
t
=
Sales
t
Sales
t 1
1.
A constant PM and ATO means constant RNOA and growth in ReOI determined only by
growth in NOA. But a constant ATO means NOA grows at the same rate as sales. One need
only forecast sales growth to get the growth rate.
Varying RNOA and Constant NOA
In this case growth in ReOI is driven by changes in RNOA. One would expect it to apply
in the case of declining RNOA, not increasing RNOA. RNOA and growth in NOA are
presumably not independent. So, for example, increasing RNOA might generate more
investment in NOA, declining RNOA less investment in NOA.
2.5. The Drivers of Free Cash Flow and Dividends
Comprehensive income is dened such that for a period, t ,
CSE
t
= CSE
t 1
+CNI
t
d
t
where d
t
is net dividends (dividends + share repurchases share issues). So, recognizing
the division of comprehensive income into operating and nancing components in (2) and
the accounting equation in (3),
NOA
t
NFO
t
= NOA
t 1
NFO
t 1
+OI
t
NFE
t
d
t
. (11)
Free cash ow
t
C
t
I
t
where C
t
is cash ow from operations and I
t
is cash ow in
investing activities. Under accrual accounting, operating income modies free cash ow
for investment and accruals:
OI
t
= C
t
I
t
+ I
t
+Operating Accruals
t
,
and net operating assets are recorded by placing the adjustments to free cash ow in the
balance sheet:
NOA
t
= I
t
+Operating Accruals
t
= OI
t
(C
t
I
t
) (12)
where indicates change (growth). As, from (11), the change in NOA is also
NOA
t
= NFO
t
+OI
t
NFE
t
d
t
,
then
NFO
t
= d
t
+NFE
t
(C
t
I
t
). (13)
Rearranging the difference equations (12) and (13),
C
t
I
t
= OI
t
NOA
t
= NOA
t 1
RNOA
t

NOA
t
NOA
t 1
1
(14)
and
d
t
= C
t
I
t
NFE
t
+NFO
t
. (15)
We have developed the ratio analysis with residual income valuation in mind. But it is
clear from (14) that the same drivers that drive ReOIRNOA and NOA growthalso
drive free cash ow, so the ratio analysis also applies to discounted cash ow analysis.
Forecasting free cash ows requires no additional information once RNOA and growth in
NOA are forecasted, as (14) is just an accounting relation. And indeed one cannot imagine
forecasting free cash ow without rst forecasting protability and the investment in NOA
that drive free cash ow.
8
If dividend discounting techniques are used to value equity, (15) shows that dividends are
also forecasted from forecasts of free cash ow drivers, along with a forecast of nancing
activities: dividends are the residual of free cash ow after servicing net debt.
2.6. Forecasting and the Analysis of Current Financial Statements
The drivers identied generate both current and future rates of return and book values. The
future drivers are the building blocks of forecasted residual earnings. Drivers are identied
in current nancial statements to forecast future drivers. Accordingly, the analysis of current
nancial statements should be guided by the predictive ability criterion: any enhancement
that improves forecasts is an innovation.
This perspective calls for an analysis that distinguishes aspects of the drivers that will
likely drive the future (persistent features) and those that relate only to the present (tran-
sitory features). And some features may persist for a short period (transitory but not pure
transitory) while some may be enduring (permanent features). This distinction is partic-
ularly important for the analysis of comprehensive income (which is required for residual
income valuation), as comprehensive income below net income under FASB Statement
No. 130 typically includes a number of transitory items.
We will refer to items judged persistent as core items and those judged transitory as
unusual items. So
OI = Core OI + Unusual Operating Income (UOI)
and
NFE = Core NFE + Unusual Financial Expense (UFE).
Tax is allocated to these components unless they are reported net of tax. With this division,
RNOA =
Core OI
NOA
+
UOI
NOA
= Core RNOA +
UOI
NOA
=
Core OI from Sales
NOA
+
Core other Items
NOA
+
UOI
NOA
.
Similarly,
NBC =
Core NFE
NFO
+
UFE
NFO
= Core NBC +
UFE
NFO
.
UFE are transitory components from realized and unrealized gains and losses on nancial
items and Core NBC is the borrowing cost (cost of debt capital) minus interest yield on
nancial assets.
The distinction between core and unusual items and sales and non-sales items can be used
to further decompose ROCE. Calculate
Core Sales PM =
Core OI from Sales
Sales
and substitute in (5):
ROCE = [Core Sales PMATO] +
Core Other Items
NOA
+
UOI
NOA
+
FLEV
Core RNOA Core NBC +

UOI
NOA

UFE
NFO
. (16)
Core RNOACore NBCis the Core SPREAD. But, if one is forecasting ReOI and valuing
accordingto(9), onlythe RNOAdecompositionis relevant. Indeed, the dismissal of nancial
items from the valuation in (9) can be justied on the grounds that any deviation of current
NBC from the cost of capital for debt (like gains and losses on nancial items) is pure
transitory and does not affect the forecast of future residual income (or expense) from
nancing activities.
9
Expression (16) does not incorporate all the drivers that have been identied. This can be
done as follows:
ROCE = MSR
(Core Sales PM
ATO
) +
Core Other Items
OA
+
UOI
OA
+(OLLEV OLSPREAD) +(FLEV SPREAD)
.
where Core Sales PM
= (Core OI from Sales + io)/Sales, ATO
= Sales/OA, FLEV
NFO/(CSE + MI) and the SPREAD is decomposed as in (16). This includes the Minority
Sharing Ratio (MSR) and the operating liability leverage drivers. But it revises the asset
turnover to ATO
which excludes operating liabilities.

Focusing on ReOI and the operating activities, an item is classied as unusual if it is
not expected to repeat in the futurelike a one-time charge or prots from discontinued
operations. But recurring items for which the expected future amount is zero are also so
classied. This is typical of items that reect uctuations in market priceslike currency
gains and losses and non-service components of pension expense that result from the quasi
mark-to-market of pension liabilities. If market prices followa randomwalk current gains
and losses due to changes in prices do not predict future gains and losses.
2.7. Commentary
The ratio analysis that has been laid out here is a scheme for using ratios in an efcient
and orderly manner in valuation analysis. It has a normative avor to it but it is not offered
as the denitive analysis. It does not suggest that a ratio not identied as an element of
the decomposition herethe current ratio, for exampleis not useful in forecasting future
residual earnings. But it does avoid the pure empirical approach (in Ou and Penman (1989)
for example) of trawling through the data without structure, and choosing ratios on the basis
of whether they worked as predictors in the past.
The reader should be clear about where the analysis comes from. The only assumption is
that the dividend discount model is an appropriate model for equity valuation. The residual
income model is merely a restatement of that model using an accounting relation (the
clean-surplus relation) between dividends, earnings and book values. The analysis builds
by recognizing further accounting relations that tie components of the nancial statements
to earnings and book values and accounting ratios to ROCE. In short, apart from assuming
the dividend discount model, the analysis is driven by the structure of the accounting, and
it is the structure of accounting that ties accounting numbers to dividends in the residual
earnings model.
The analysis eschews modeling of the economics. One might model how residual income
behaves in different economic circumstances (under competition, monopoly, regulation,
for example) but residual income is an accounting calculation. Residual income methods
work, whatever the economic circumstances. Modigliani and Miller (economic) concepts
are referred to but the analysis leaves the user with a choice as to whether to embrace those
concepts.
10
Economic factors, of course, determine value. We speakhere of nancial statement drivers,
but these drivers are driven by the economics of the rm. The analysis here identies the
accounting drivers of residual earnings that attach to economic factors. Indeed, analysis
of nancial statements directs how to analyze business activity and to translate economic
factors into terms (accounting drivers) that forecast residual earnings and lead to a valuation.
Many of the ratios are familiar and the basic structure follows the familiar DuPont scheme.
But there are innovations:
Ratio analysis is integrated with valuation analysis, giving substance to fundamental

analysis.
The analysis is of comprehensive income. Components of other comprehensive income

and dirty-surplus income in the balance sheet in the U.S. (and items in the Statement of
Recognized Gains and Losses and Reconciliation of Shareholder Funds Statements in the
UK) are included in ratios. Residual earnings valuation techniques require the forecasting
of comprehensive income, otherwise value is lost.
The analysis takes a forecasting perspective. Ratio analysis is seen as an analysis of current
nancial statements but also as an analysis of future residual earnings. Forecasted ratios
are the buildingblocks of forecastedresidual earnings. Current ratios forecast future ratios.
The analysis of current nancial statements is guided by the principle of predictive ability.
So transitory and core aspects of ratios are identied (in principle).
The decomposition leads to parsimony in analysis. Ratios further down the hierarchy are
utilized only if they provide more information than those higher up. Ratios that involve
nancing activities are ignored if nancial items are at their fair value on the balance
sheet. RNOA and growth forecasts can be simplied if components are constant, so the
analyst can focus on the key drivers that will affect the forecast.
Protability analysis is at the heart of the analysis, but this is complemented with an
analysis of growth. And growth is given explicit expression.
There is an extended analysis of drivers of protability beyond the standard analysis.
Minority interest share in accounting value is accommodated.
There is a clear distinction between operating and nancing items. This is done by apply-
ing clean-surplus accounting, not just between the income statement and balance sheet
totals, but between operating and nancing totals on the income statement and balance
sheet: NOA are identied in the balance sheet to match to operating income, and NFO
are separated to match to net nancial expense, to yield clean measures of operating
protability and borrowing costs.
Financial leverage is redened from the traditional Debt/Equity ratio.
An operating liability leverage driver is identied and an analysis of favorable and unfa-
vorable operating liability leverage is given.
If nancing assets and liabilities are measured at market value, the analysis should focus
on the operating activities to complete the valuation. The last three points are important to
the discovery of operating protability. Some calculations of operating protability mingle
operating and nancing items. Acommon calculation is return on total assets (often referred
to as return on assets, ROA):
ROTA =
NI +Interest Expense (1 t ) +MI share of income
Total Assets
(17)
where t is the taxrate. This oftenignores items incomprehensive income (bystartingwithnet
income) but also includes interest income in the numerator and nancial assets in the denom-
inator. Further, operating liabilities are not subtracted fromthe denominator. The analysis in
the paper makes the appropriate separation, yielding the RNOA measure of operating prof-
itability. The identication of operating liabilities (as distinct from nancing), leads to the
notion of operating liability leverage and the analysis gives an explicit expression (7) for it.
The reduced analysis of Section 2.3 aims at focusing on the aspects of the business that
generate value, the operations. Clearly there is a question of denition (of operating and
nancing activities) but we have in mind that operations are carried out to make money,
as distinct from the zero-residual-earnings activities (tax issues aside) involved in nancing
these activities. So, buying and selling bonds at market price to raise cash for an industrial
rm is a nancing activity, but buying and selling bonds for a bond trader is an operating
activity. Knowing the business is important to the identication. The identication can only
be made if operatingandnancingactivities are separable. So, if a rmholds debt ina foreign
currency to hedge against exchange rate losses from operations, the separation cannot
be made. And, if disclosure is insufcient to make the distinction, the separation cannot
be made. The principle of clean surplus accounting requires that operating and nancing
income, separately identied in the income statement, must be matched with operating
and nancing net assets separately identied in the balance sheet. Otherwise measures of
operating protability and net borrowing cost that involve the matching are dirty.
If operating and nancing items cannot be identied (for lack of disclosure), the reduced
analysis is not feasible and the analyst works with forecasting residual earnings, as in
Section 1, rather than residual operating earnings. This requires an analysis of the nancing
activities, as in Sections 2.1 and 2.2. And, it involves adjustments to the cost of equity
capital for continually changing forecasted leverage.
3. Documentation
There are two parts to the documentation, a cross-sectional analysis and a time series
analysis. The cross-sectional analysis gives typical numbers for the ratios in the data. This
is of particular help for ratios like RNOA, OLLEV, FLEV and the core ratios which may
be unfamiliar or which are dened differently from standard texts. The time series analysis
documents how ratios typically evolve over time. With the view of using current drivers
as predictors of future residual earnings drivers, the time series analysis documents the
transition from current drivers to future drivers.
The documentation is for rms using U.S. GAAP. It covers NYSE and AMEXrms listed
on the combined COMPUSTAT (Industry and Research) les for the 37 years from 1963
to 1999. These are relatively well-established rms. Non-surviving rms are included. The
appendix explains how ratios are calculated from COMPUSTAT items. Some calculations
are hampered by insufcient disclosure in the nancial statements, to distinguish operating
and nancing items and core and unusual items, for example. But our analysis is also
somewhat restricted by lack of data on COMPUSTAT and these difculties are discussed
in the appendix.
Many ratios can have extreme values, usually due to very small denominators. So in most
cases, median ratios are presented as representative numbers. Where means and standard
deviations are reported, they exclude the upper and lower one percent of the distribution.
Ratios with negative denominators are also excluded.
11
Balance sheet numbers are averages
of beginning and ending amounts. All income numbers are after tax, with the appropriate
tax allocation at all points. For residual earnings and residual operating earnings, we set
the cost of capital at the current one year treasury rate plus 6%, the conjectured equity risk
premium. We do not attempt to distinguish levered and unlevered costs of capital. This
reects our uncertainty about the appropriate risk premium more than anything else; in any
case our focus is on the accounting numbers, not the cost of capital.
Before providing the documentation, we should indicate that we began the empirical
analysis by attempting to estimate multivariate models to forecast residual operating in-
come, RNOA, and growth in NOA from the pooled cross-section and time-series data.
With parsimony in mind, these models were estimated by including ratios in the hierar-
chical order of the decomposition so that ratios were only introduced if they had explana-
tory power beyond higher level ratios under which they nest. The analysis produced large
t -statistics and reasonable R-square values in estimation, but the models performed poorly
in prediction out of sample. This was the experience in Ou and Penman (1995) also. We
became convinced that coefcients in these models are not stable across rms and time,
lending credence to the conjecture that nancial statement analysis is contextual. Accord-
ingly the documentation here is more descriptive, designed to identify empirical regularities
and provide general benchmarks as a point of departure for the contextual analysis of indi-
vidual rms. As it happens, we will show that the relationship between current and future
drivers is non-linear, so pooled, linear models are not likely to work well. (None of this is
to imply that robust, parametric, predictive models cannot be estimated; but that estimation
calls for careful econometrics and a careful partitioning of the data.)
3.1. Cross-sectional Analysis
Typical Ratios
Table 1 summarizes the mean, median and other aspects of the distribution of ratios pooled
over all rms and all years, 19631999. The rst panel gives the main drivers of the ROCE
component of residual earnings. The second panel gives ratios that help isolate core prof-
itability and measures that drive the growth of residual earnings.
The minority sharing ratio (MSR) is close to one for a large section of rms, so ROCE is
typically a good approximation for ROTCE. Median ROCE (12.2%) is, interestingly, close
to, or perhaps a little higher than what is normally assumed as the equity cost of capital:
Ibbotson and Sinqueeld (1983) calculate the historical average return to equity at about
12.5%at a point about half way through our sample period, although this is often claimed to
be too high because it reects an ex post successful stock market (See Brown, Goetzmann
and Ross (1995)).
RNOA is higher at the mean and median than the traditional ROTA, and has considerably
higher variation. The distribution of the difference between the two demonstrates that the
clean distinction between operating and nancing items and the adjustment for operating
liabilities can have a signicant effect: for almost 50% of rm-years the absolute difference
is greater than 3%. The median RNOA of 10.0% compares with a median ROTA of 6.8%
and is closer to what we typically think of as an average business return. The numbers for
ROTA (or return on assets) reported in texts and in the business press often seem too low
and we suggest that this is due to poor measurement. Median operating liability leverage
(OLLEV) and median operating liability spread (OLSPREAD) are positive (0.35 and 3.4%,
respectively), so typically operating liability leverage is used favorably, contributing to the
difference between RNOA and ROTA.
The distinction between operating and nancing assets and liabilities changes the nancial
leverage measure, FLEV, from the traditional Debt/Equity ratio measured as (Total Lia-
bilities + Preferred Stock)/Common Equity. Median FLEV is 0.40, compared to 1.19 for
the Debt/Equity ratio. This is because FLEV recognizes only indebtedness from nancing
activities and also recognizes that debt held (as assets) effectively defeases debt owed. So
for about 20% of rms nancial leverage is negative; these rms are net holders of nancial
assets rather than net issuers.
Net borrowing costs (NBC) in the table are after tax. Adjusting the median 5.2% for
tax rates (see appendix), the before-tax rate is typical of corporate borrowing rates. The
variation in borrowing cost is probably due to variation in borrowing rates but also to
recognition of (transitory) realized and unrealized gains and losses on nancial items. The
SPREAD over the net borrowing cost is positive at the median but negative for about 30%
of rm-year observations. Median ROCE (12.2%) is higher than median RNOA (10.0%),
indicating that typically the on-average positive leverage combines with positive spread to
lever ROCE favorably.
The standard DuPont prot margin (PM) and asset turnover (ATO) which drive RNOA
are given in the last two columns of Panel A of Table 1. The rst eight ratios in Panel B
(from Sales PM to Core SPREAD) are additional ratios that extend the standard analysis of
ROCEin (4a) to that in (16). The Sales PMshifts the PMslightly lowerfrom5.5%to 5.4%
at the median, more so in the positive tailbecause of the exclusion of Other Items from
the numerator of Sales PM. Other Items/NOA is typically small; only about 20% of rms
report equity earnings in subsidiaries. But the identication of unusual operating items has
a larger effect, particularly away fromthe median. Comparison of Core Sales PMwith Sales
PM and of Core RNOA with RNOA reects the effect of these unusual items. This effect
is understated here since COMPUSTAT does not give enough nancial statement detail to
identify all unusual items. Further unusual items, such as strikes and unusual orders, can be
discovered from footnotes and the management discussion and analysis. A more thorough
analysis of unusual items will rene the Core RNOA further.
NBC and SPREAD in the table are presented with qualications. Realized gains and
losses on debt are not identied by COMPUSTAT and hence are not included in NBC.
Unrealized gains on long-term nancial assets and unrealized gains and losses on short-
term nancial assets have been recognized only since FASB Statement No. 115 became
effective in 1994.
12
Prior to 1994 (when only unrealized losses on long-termnancial assets
were recognized), our measured UFE was zero for almost all rms. Thus the table presents
Core NBC, UFE/NFO and Core SPREAD for 19941999 only. Indeed our identication of
unusual nancial items is a response to Statement No. 115 and the distribution here gives
an indication of how accommodation of this statement affects the numbers.
The remaining columns of Panel Bgive growth rates and free cash ow. Continuing value
calculations require growth rates for either residual earnings (RE) or residual operating
income (ReOI). Panel B gives the distribution of annual growth rates during 19631999. If
ROCE is constant, growth in residual earnings is driven by growth in CSE, and if RNOA is
constant, growth in ReOI is driven by growth in NOA. So growth rates in CSE and NOAare
also given. Median growth in CSE is 9.0% and median growth in NOA is 8.9%. These are,
however, annual realized growth rates and the continuing value calculation requires long-run
expected growth rates. So the relevant question is howgrowth rates settle down in the long-
run, and we return to this issue later.
13
Protability and growth in NOAyielded a median free
cash ow of 2.1% of NOA and, for about 40% of rm-years, free cash ow was negative.
Decomposition of the primary drivers is only of use if it provides more information.
Table 2 gives a matrix of Spearman correlations for the ratios summarized in Table 1. We
leave it for the readers inspection.
Typical Ratios Over Time
The ratios in Table 1 are typical of the period. But they give no indication of the variation
and trends over time that is helpful for prediction. Figures 1a1f trace median values over
the 37-year period. With forecasting in mind, one might look at these gures as a basis for
projecting to the future (beyond 1999). Permanent trends might be extrapolated. More
recent numbers might be given more weight but might also be interpreted against any
historical tendency to revert to central or typical values.
Figure 1a plots median ROCE against the one-year treasury yield (the risk-free rate)
and our estimate of the cost of equity capital (the treasury yield plus 6%). We noted in
Table 1 that the grand median ROCE of 12.2% looked like the equity return and Figure 1a
indicates this is consistently so; median ROCE is greater than the risk-free rate in all years
except 1982 and tracks the cost of capital (somewhat surprisingly).
14
But ROCE is less than
the assumed cost of equity (implying negative residual earnings) in a majority of the years,
suggesting that the cost of capital estimate is too high. If anything, one expects ROCE to be
greater than the cost of capital because conservative accounting, that keeps book values low
and ROCE high, is said to be practiced. Interestingly, Claus and Thomas (2001), OHanlon
and Steele (1998) and Gebhardt, Lee and Swaminathan (2000) impute lower equity risk
premiums than the 6% from forecasted residual earnings in the U.S. and actual residual
earnings in the UK. In any case, there is a central tendency in the economy-wide ROCE
in Figure 1a which should be noted for forecasting: it moves around its grand median and to
some extent, tracks the cost of capital. Results are similar with the ve-year treasury yield
as the risk-free rate.
Figure 1b plots median ROCE and its two drivers, RNOA and NBC. The (after-tax) NBC
can be compared with the (pre-tax) treasury yield in Figure 1a. Periods where interest rates
Figure 1. (a) Median ROCE, one-year treasury rate and the assumed cost of equity capital. (b) Median ROCE,
RNOA and NBC over time. (c) Median FLEV, OLLEV and debt/equity over time. (d) Median ROTA and RNOA
over time. (e) Median RNOAand its components over time. (f ) Median core RNOAand its components over time.
Figure 1. (Continued )
move away fromNBCare those in which nancial obligations are less likely to be at market
value. ROCE is consistently higher than RNOAindicating a consistently favorable nancial
leverage effect. Median RNOA is consistently a positive spread above net borrowing cost.
Figure 1c plots median FLEV that, with the SPREAD, produces the leverage effect, and
contrasts it to the Debt/Equity ratio. Figure 1c also plots the OLLEV that levers the RNOA.
FLEV and OLLEV are, year-to-year, quite stable. OLLEV has a slight positive trend;
the recognition of employment benet liabilities in the late 1980s and 1990s presumably
contribute. The median Debt/Equity ratio increased more than FLEVover the years, induced
by an increase in operating liabilities. The Debt/Equity ratio differs from FLEV also in the
treatment of nancial assets. Cash from operations can be used to buy the rms own debt
(which affects the Debt/Equity ratio) but can also be used to buy nancial assets, others
debt (which does not affect the Debt/Equity ratio). In both cases the rm is buying debt,
engaging in a nancing transaction that reduces its net debt but does not affect the operations.
The FLEV measure, based on a separation of nancing and operating activities, captures
both cases; the Debt/Equity ratio does not.
Figure 1d plots median ROTA and RNOA and Figure 1e plots core and unusual com-
ponents of RNOA. ROTA consistently understates RNOA but the two do move together.
Core RNOA is consistently lower than RNOA, except for some years in the 1990s when,
one conjectures, the many restructurings and accounting charges produced large negative
unusual charges.
Figure 1f plots median Core RNOA again but also plots its drivers, Core Sales PM, Other
Items/NOA and ATO. In the early part of the sample period, changes in economy-wide
Core RNOA seem to have been driven more by changes in ATO than changes in Core Sales
PM. But during the last 15 years Core Sales PM has driven the Core RNOA, with median
ATO reasonably constant.
Interaction between Ratios
The extended decomposition in (16) gives the drivers of ROCE and shows how they ag-
gregate. But there may be interactions: a certain level of one driver may imply a certain
level for another. Here we examine the data for these interactions. They are depicted in
Figures 2a2d. The gures plot joint values of median ratios for 244 three-digit SIC indus-
try groups over 19631999. Plots were also made for just the more recent 19901999 period
with similar results.
15
These plots should be read in conjunction with the correlations in
Table 2.
FLEV and SPREAD. Financial leverage levers ROCE relative to RNOA, as in (4), and
the amount of leverage depends on both FLEV and SPREAD. As the effect of leverage
(favorable or unfavorable) depends on the sign of the SPREAD, FLEV may be set by
management after contemplating the SPREAD the rm will generate. One might expect a
positive relationship between FLEVand SPREAD: a rmborrows more (to lever up ROCE)
only if it can maintain high SPREAD which is less likely to turn unfavorable. But some
argue that nancing is irrelevant. Positive leverage generates higher anticipated ROCE, but
increases the risk of lower protability. So higher anticipated residual earnings from an
Figure 2. (a) Median FLEV on median SPREAD. (b) Median OLLEV on median OLSPREAD. (c) Median core
sales PM on median ATO. (d) Median RNOA on median growth in NOA.
increase in anticipated ROCE are exactly offset in present value form by an increase in the
cost of equity capital. Accordingly management may choose leverage for reasons other than
increasing protability of the equity.
Figure 2a plots FLEV on SPREAD. The leverage effect is given by FLEV*SPREAD
and we can think of tting iso-leverage effect curves through the plots. The relationship
is, surprisingly, negative. The Spearman rank correlation between FLEV and SPREAD
is 0.38, and between FLEV and RNOA is 0.42. Perhaps a high median RNOA or
SPREAD is the reward to business risk and rms with high business risk choose to have
lower nancing risk. Also, higher nancial leverage presumably results in higher borrowing
costs, reducing the SPREAD. And perhaps protable rms generate a lot of cash which
they use to reduce leverage.
OLLEV and OLSPREAD. Operating liability leverage levers RNOA relative to ROOA, as
in (7), and the amount of leverage depends on both OLLEV and OLSPREAD. In a similar
way to Figure 2a, Figure 2b plots the interaction between OLLEV and OLSPREAD. The
rank correlation between the two is 0.15, and the rank correlation between OLLEV and
RNOA is 0.28, indicating operating liability leverage works favorably, on average.
PM and ATO. The DuPont decomposition recognizes that RNOA = PMATO and it is
commonly recognized that rms can generate the same RNOA with different combinations
of margins and turnovers. Figure 2c replicates the gure in Selling and Stickney (1989) that
is often displayed in texts, but the prot margin is the rened Core Sales PM and ATO here
is based on NOA which incorporates operating liabilities. Iso-RNOA lines tted through
these plots are convex and downward sloping. The rank correlation between Core Sales PM
and ATO is 0.40.
RNOA and Growth in NOA. RNOAand Growth in NOAcombine to growresidual operat-
ing income. One might expect rms that generate higher RNOA to grow their net operating
assets. But growth in NOA reduces RNOA if the accounting is conservative. Figure 2d
shows that median RNOA and Growth in NOA are positively related in the sample period.
The rank correlation between the two is 0.24.
The full set of correlations in Table 2 are at the rmlevel, not the industry level. Inspection
will reveal some further relationships. For example, Core NBC is only slightly negatively
correlated with RNOA (0.08) but is positively correlated with FLEV (0.28); FLEV and
OLLEV are negatively correlated (0.26); and OLLEV is negatively correlated with Core
Sales PM (without the recognition of imputed interest expense), but positively correlated
with ATO, and overall OLLEV is positively correlated with RNOA.
3.2. Time-Series Analysis
This section documents the typical evolution of ratios over time. Ratios were identied in
Section 2 as drivers of future residual earnings. With a view to forecasting, the analyst is
not primarily concerned with current drivers, but with how current drivers will transition
to the future. Will they persist or will they decay? If ratios decay, what will be their future
level?
Of particular interest is the question of whether drivers tend to converge to typical values
overtime. As with all valuation methods, a residual earnings valuation can be made from
forecasts for truncated forecast horizons if attributes settle down to permanent levels
within the horizon.
16
If they do, continuing values can be calculated. Practical analysis
typically makes assumptions about continuing values, as does academic research using
analysts forecasts (which are typically made only for two to ve years).
17
Do residual
earnings and their drivers typically settle down in the way assumed? And, if so, what is
the typical form of the continuing value calculation?
We examine these issues by reference to the displays in Figures 3, 4, 5 and 6, along with
simple rank correlation measures. The displays are based on ranking a given measure in a
base year, year 0, forming 10 portfolios from the rankings, and then tracking median values
for each portfolio for the following ve years. The ranking is done seven times, in 1964,
1969, 1974, 1979, 1984, 1989 and 1994. For each of these years, portfolio medians are
calculated for the ranking year and for the following ve years. At end of the ve years
the ranking is done again for the next ve years, and so on until 1994. The gures give the
mean of portfolio medians over the seven sets of calculations. The patterns depicted are quite
robust over the seven time periods, however. The displays are offered with one caveat. Firms
in the base year that do not survive are not in the calculations in years after they drop out.
Thus the expost averages may be biased estimates of exante amounts for going concerns.
The displays are to be read in conjunction with Table 3.That table gives the Spearman rank
correlations between ratios and their t -lagged values (t = 1, 2, . . . , 5), calculated using all
rm-year observations (19631999). It also gives the ratio of the variance of the portfolio
measures in year t (t = 1, 2, . . . , 5) with that in the base year (year 0), as an indication (at
the portfolio level) of the speed of conversion toward a common amount. The correlations
are analogous to R
2
in a linear regression and the variance ratios are analogous to a slope
coefcient.
Figure 3. (a) Evolution of RE over time. (b) Evolution of ReOI over time. (c) Evolution of FLEV over time.
Figure 4. (a) Evolution of ROCE over time. (b) Evolution of RNOA over time. (c) Evolution of growth in CSE
over time. (d) Evolution of growth in NOA over time.
Residual Earnings (RE), Residual Operating Income (ReOI) and FLEV. Figures 3a and
3b give the average behavior of median RE and ReOI over ve years from the base year.
Both are per dollar of average book value (CSE and NOA respectively) in the base year.
Both are calculated with a cost of capital equal to the one-year risk-free rate at the beginning
of the year plus 6%, but the patterns are similar when the cost of capital for the subsequent
years is the risk-free rate for the future year that is implied in the term structure at the end
of the base year, plus 6%.
The followingobservations are made for bothREandReOI. First, current residual earnings
forecast future residual earnings, not only in the immediate future but ve years ahead. High
residual earnings rms (in the cross section) tend to have high residual earnings later and low
residual earnings rms tend to have lowresidual earnings later. Indeed, the rank correlations
in Table 3 between RE in year 0 and RE in years 1, 2, 3, 4 and 5 for individual rms are 0.55,
0.37, 0.25, 0.20, and 0.18, respectively. The corresponding rank correlations for ReOI are
0.65, 0.50, 0.40, 0.35, and 0.33. Second, in both the gures, the decaying rank correlations,
and the decaying variance ratios indicate residual earnings tend to converge to central values,
with the more extreme becoming more typical over time. But third, permanent levels are
not zero for a number of portfolios. It appears that continuing values of the type CV2 and
CV02 (or CV3 and CV03) are typical, not CV1 or CV01. A nonzero permanent level of
residual earnings can be explained by persistent nonzero net present value investing or by
conservative or liberal accounting. Fourth, long-run growth in residual earnings is slightly
positive, suggesting that continuing value calculations of the type CV3 and CV03 may
improve upon CV2 and CV02.
One other aspect of Figures 3a and 3b warrants mention. The majority of rms have
negative RE or ReOI by the fth year. Re and ReOI are determined in part by the cost of
capital. With conservative accounting practiced on average, one expects long-run RE and
ReOI to be, on average, positive and ROCE and RNOAto be greater than the cost of capital.
Figure 5. (a) Evolution of core sales PM over time. (b) Evolution of other core items/NOA over time.
(c) Evolution of UOI/NOA over time. (d) Evolution of ATO over time.
If this is the case, the displays indicate that the typically assumed equity risk premium of
6% is too high. As mentioned earlier, risk premiums estimated using the residual earnings
model are less than 6%.
Financial leverage drives the difference between RE and ReOI so Figure 3c shows how
FLEV changes overtime. FLEV is fairly constant for all portfolios except that with the
highest FLEV. A temporarily high FLEV can produce temporarily high RE (through the
leverage effect on ROCE) even when RNOA is normal. But mean reversion in FLEV
forecasts that unusually high RE, so induced, will become more typical. Valuation by
forecastingReOI avoids these considerations andalsoavoids forecastingchanges indiscount
rates due to forecasted changes in leverage.
ROCE, RNOA, Growth in CSEand Growth in NOA. Residual earnings measures are driven
by rates of return and growth in book value. Figures 4a and 4b trace portfolio medians of
ROCE and RNOA and Figures 4c and 4d trace portfolio medians of annual growth rates in
CSE and NOA. The mean reversion in ROCE in Figure 4a is well documented (in Freeman,
Ohlson and Penman (1982), Penman (1991) and Fama and French (2000)). Asimilar central
tendency is evident for RNOA. So the mean reversion in RE and ReOI is driven in part
by that in the two rates of return. However, Figures 4c and 4d indicate that growth in
book value also decays toward economy-wide levels, driving RE and ReOI toward central
values.
18
Core Sales PM, Other Items, Unusual Operating Items and ATO. Figures 5a, b, c and d
plot the drivers of RNOA, with Other Core Items and Unusual Operating Items deated
by NOA.
19
The division of protability into Core Sales PM, Other and Unusual Items
distinguishes components with different persistence. Differences in Core Sales PM per-
petuate. And, with the exception of the highest ATO group, asset turnovers remain fairly
constant.
Figure 6. (a) Evolution of sales growth rate PM over time. (b) Evolution of change in core sales prot margin
over time. (c) Evolution of change in ATO over time.
As indicated in Section 2.4, forecasting that prot margins and ATO will be constants
simplies the forecast of growth: the growth rate in ReOI for a continuing value calculation
is then determined solely by forecasting growth in sales.
Changes in Sales, Core Sales PMand ATO. The Figure 5 graphs are for portfolios formed
from the cross section. Figure 2c indicates systematic differences in prot margins and
turnovers across industries, so the differences in the long-run prot margins and ATOs in
Figure 5 are probably due to permanent differences in technology and cost structure across
Table 3. Summary measures for the evolution of ratios.
Year t Relative to Base Year 1 2 3 4 5
RE corr
t
0.551 0.372 0.257 0.200 0.176
var
t
/var
0
0.441 0.241 0.165 0.188 0.209
ReOI corr
t
0.653 0.496 0.397 0.349 0.327
var
t
/var
0
0.705 0.549 0.501 0.473 0.568
FLEV corr
t
0.937 0.846 0.781 0.735 0.700
var
t
/var
0
0.900 0.772 0.674 0.583 0.526
ROCE corr
t
0.618 0.425 0.311 0.253 0.230
var
t
/var
0
0.372 0.183 0.100 0.080 0.067
RNOA corr
t
0.665 0.485 0.377 0.320 0.298
var
t
/var
0
0.534 0.318 0.204 0.155 0.127
Growth in CSE corr
t
0.659 0.307 0.190 0.129 0.105
var
t
/var
0
0.408 0.088 0.038 0.024 0.020
Growth in NOA corr
t
0.584 0.165 0.074 0.051 0.051
var
t
/var
0
0.318 0.025 0.011 0.008 0.006
Core Sales PM corr
t
0.828 0.725 0.664 0.628 0.605
var
t
/var
0
0.808 0.691 0.617 0.555 0.534
Other Core Items/NOA corr
t
0.663 0.533 0.451 0.406 0.367
var
t
/var
0
0.645 0.422 0.311 0.243 0.195
Unusual Items/NOA corr
t
0.376 0.284 0.223 0.212 0.189
var
t
/var
0
0.128 0.058 0.031 0.037 0.020
ATO corr
t
0.947 0.894 0.857 0.830 0.808
var
t
/var
0
0.850 0.739 0.720 0.662 0.602
Growth in Sales corr
t
0.309 0.115 0.103 0.106 0.115
var
t
/var
0
0.073 0.012 0.011 0.008 0.007
Core Sales PM corr
t
0.039 0.083 0.059 0.043 0.012
var
t
/var
0
0.013 0.010 0.003 0.001 0.001
ATO corr
t
0.166 0.100 0.082 0.039 0.009
var
t
/var
0
0.031 0.024 0.010 0.003 0.003
Corr
t
is the Spearman rank correlation between each ratio and its t -lagged value calculated using all rm-year
observations (19631999). The variance ratio is based on the portfolio analysis. This analysis is done by forming
ten portfolios from a ranking on a ratio in a base year, and calculating the median ratio for each portfolio in the
base year and each of the subsequent ve years. (The base years are 1964, 1969, 1974, 1979, 1984, 1989 and
1994.) Next, means of portfolios medians over the seven replications are calculated (Figures 3a6c plot these
means). var
t
is the variance of the mean of portfolios medians in year t . Thus, var
t
/var
0
measures the speed of
mean reversion.
industries. So Figures 6a, b and c picture the evolution of changes in sales, Core Sales PM
and ATO.
The patterns indicate that unusual charges are indeed transitory. Figure 6b also indicates
that Core Sales PM can contain further transitory income (after isolating unusual items)
and this is detected by looking at changes.
20
Similarly, changes in ATO are mean reverting,
and rapidly so.
21
Table 3 and the supporting displays incorporate all rms that survive to each forward
year. If the survival rate differs over the portfolios or the reason for non-survival differs,
these expost representations may be biased depictions of exante patterns. Correcting for any
effect of attrition is problematical. One might include expost liquidating earnings for rms
that drop out but exante forecasts are usually made for going concerns (without anticipation
of the effects of termination).
Panel A of Table 4 gives the average survival rates for the 10 residual earnings portfolios
in Figure 3a. The numbers in the table are averages over the seven replications (19641969,
Table 4. Percentage survivorship rates for residual earnings portfolios.
Year t Relative to Base Year: 1 2 3 4 5
Panel A: Percentage survivorship rates
Portfolio:
10 (high RE) 97.84 94.66 91.74 89.30 85.45
9 97.57 95.49 93.59 91.01 87.41
8 97.38 95.52 93.32 90.77 87.61
7 96.75 94.77 92.29 89.73 86.47
6 97.33 94.98 91.95 89.83 85.76
5 97.46 94.78 92.50 89.25 85.57
4 97.67 95.79 93.34 89.79 85.47
3 96.77 94.36 91.13 87.39 82.17
2 95.80 91.97 88.27 84.74 79.63
1 (low RE) 94.13 89.90 84.83 80.58 75.94
Panel B: Percentage of identied terminations that are due to merger or exchange
Portfolio:
10 (high RE) 78.00 87.33 88.71 89.93 88.73
9 93.33 95.26 96.55 96.55 96.05
8 100.00 96.00 96.58 97.45 97.56
7 96.67 96.47 95.82 95.84 95.99
6 100.00 97.90 98.02 97.38 97.44
5 90.00 92.46 93.81 95.98 96.37
4 90.00 95.00 96.56 96.06 96.79
3 95.00 93.75 94.89 95.41 95.32
2 77.50 90.57 93.14 92.77 89.44
1 (low RE) 66.57 71.23 74.20 75.78 72.73
Panel A reports average (over the seven replications: 1964 1969, . . . , 1994 1999) of the per-
centage of rms in each of the 10 residual earnings (RE) portfolios that survived in the respective
year after the base year.
The main four reasons for non-surviving are mergers, exchange of stock, liquidation and delist-
ing. Panel B gives the percentage of non-surviving rms due to a merger or exchange out of all
non-surviving rms for which the reason of non-survival is reported in CRSP.
19691974, . . . , 19941999) of the percentage of rms in each of the 10 portfolios in the
base year that survived in each of the subsequent 5 years. Survivorship rates are similar
across all portfolios except the bottom two (with the lowest RE). Patterns for all groups
are conditional upon survival, but, for these two groups, the forecaster must be particularly
concerned with forecasting survivorship as well as forecasting drivers. The main reasons for
non-surviving are mergers, exchange of stock, liquidation and delisting. Panel B of Table 4
gives the percentage of terminations due to merger or exchange, as indicated by CRSP.
Myers (1999) indicates that mergers and exchanges result in considerably higher payoffs
than liquidation or delisting and, indeed, higher payoffs than those of surviving rms. The
lower residual earnings portfolios have fewer mergers and exchanges (and more delistings
and liquidations). And the fade diagrams, like Figure 3a, do not capture the higher payoffs
to mergers and exchanges.
4. Conclusion
This paper has laid out a structured nancial statement analysis that facilitates forecasting
and valuation. The analysis involves an analysis of protability and an analysis of growth.
The analysis of protability extends the traditional analysis, the analysis of growth comple-
ments it: protability and growth drive equity values. The analysis is guided by the residual
earnings valuation model but is appropriate for forecasting free cash ows and dividends if
other valuation approaches are adopted.
Section 2.7 summarizes the major features of the scheme. The overall perspective is one
of forecasting (and, fromforecasts, valuation) so nancial statement analysis is depicted, as
a matter of rst order, as an analysis of the future. Ratios are identied as drivers of future
residual earnings, free cash ow and dividends. Ratios in current nancial statements are
then viewed as information to forecast the future drivers.
The paper documents typical values for ratios and typical patterns for ratios over time.
The broad general descriptions are benchmarks that establish priors for forecasting. They
are input to a more contextual analysis (of industries and rms) and a more sophisticated
econometric forecasting analysis. The overall feature of the accounting data is central
tendencyconvergence of drivers to a common level in the cross sectionwhich bodes
well for calculating continuing values with standard valuation techniques.
The analysis does not deal withuncertaintyinforecasting. This is incorporatedinvaluation
models through the discount rate (or discounts from expected values) and one conjectures
that nancial statement analysis is also relevant for determining the discount rate.
Accounting research is concerned with assessing the information content of nan-
cial statements, presumably with a view to developing concrete products that can be
used in practice. The practical signicance of this paper should be obvious. But the anal-
ysis also guides future research. For forecasting research that applies the predictive
ability criterion to evaluate information content, the analysis indicates what is to be fore-
casted if the researcher has valuation in mind. It also identies the accounting informa-
tion that drives the forecast and the form it might take. The empirical analysis here is
very much at the descriptive level but, with thought, may lead to more formal, parsimo-
nious forecast modeling which brings more sophisticated econometrics to the task. Sim-
ilarly, the analysis aids in the development of accounting-based valuation models that
go beyond earnings and book values. And it guides the specication of capital mar-
kets regressions that conduct empirical tests of information content and accounting-based
valuation.
Appendix: Notation and Variables Measurement
This appendix describes how the variables are measured. Since perfect identication of
operating and nancing items requires data that are not available on Compustat, some of
the variables are measured with error. When applicable, we discuss measurement issues
below the variables denition.
Financial Assets (FA) = cash and short term investments (Compustat #1)
plus investments and advances-other (Compustat #32).
Companies hold cash on a regular basis partially because they cannot perfectly synchro-
nize cash receipts and disbursements. Thus, a portion of cash and short-term investments
is an operating asset. But we cannot separate the operating portion of cash from the total of
cash and short-term investments, and we cannot separate interest income on operating cash
from total interest income. So we classify all cash as a nancial asset.
There are generally two types of investment securities: debt and equity. Investments in debt
securities are nancial assets, but investments in equity securities are usually investments in
the operations of afliated companies, and thus are operating assets. Compustat reports long-
term investments in debt securities together with long-term investments in equity securities
(other than those accounted for by using the equity method). Since most investments (other
than those accounted for using the equity method) are in debt securities, we classify the
item investments and advancesother (#32) as part of nancial assets.
Another measurement issue is that investments and advancesother (#32) includes
long-term receivables that are related to operating activities. However, receivables are usu-
ally carried close to their value even when they relate to operating activities (e.g., long-term
lease receivables). More importantly, interest income onall receivables is includedininterest
income, which is classied as nancial income (see below).
Operating Assets (OA) = Total Assets (TA, Compustat #6) minus Financial Assets (FA).
Financial Obligations (FO) = debt in current liabilities (#34) plus long term debt (#9)
plus preferred stock (#130) minus preferred treasury stock (#227) and plus preferred
dividends in arrears (#242).
Net Financial Obligations (NFO) = Financial Obligations (FO) minus Financial Assets
(FA).
Common Equity (CSE) =common equity (#60) plus preferred treasury stock (#227) minus
preferred dividends in arrears (#242).
Net Operating Assets (NOA) = Net Financial Obligations (NFO) plus Common Equity
(CSE) and plus Minority Interest (MI, #38)
Operating Liabilities (OL) = Operating Assets (OA) minus Net Operating Assets (NOA).
Core Net Financial Expense (Core NFE) =after tax interest expense (#15 (1 marginal
tax rate)) plus preferred dividends (#19) and minus after tax interest income (#62
(1 marginal tax rate)).
Unusual Financial Expense (UFE) =lag marketable securities adjustment (lag #238) minus
marketable securities adjustment (#238).
Due to data problems we cannot include gain/loss on early extinguishment of debt, a
component of UFE, in our measure of UFE. This item is classied as an extraordinary
item and Compustat provides only the total of extraordinary items with no information on
their nature. So we include all extraordinary items in operating income. In any case, our
classication excludes extraordinary items from core income.
Another data issue withthe measurement of UFEis that cumulative translationadjustment,
a dirty surplus equity account, reects unrealized currency translation gains and losses not
only on operating items, but also on nancial assets and liabilities. COMPUSTAT(and often
the nancial reports) does not provide enough details to separate the two. Since NOA are
usually larger than NFO, we include this income item in operating income (see below).
Net Financial Expense (NFE) = Core Net Financial Expense (Core NFE) plus Unusual
Financial Expense (UFE).
Clean Surplus Adjustments to net income (CSA) =marketable securities adjustment (#238)
minus lag marketable securities adjustment (lag #238) plus cumulative translation ad-
justment (#230) and minus lag cumulative translation adjustment (lag #230).
CSAshould include all items that change common equity and bypass the income statement
(dirty surplus items) other than net dividends. Compustat provides information on the
cumulative balance of two dirty surplus items: translation gains and losses (#230) and
unrealized gains and losses on nancial items (#238). Thus, CSA may be estimated as the
change in these items during the year. These two items, although the more common and
material dirty surplus items, are not the only ones.
An alternative approach to calculating CSA is by analyzing the change in retained earn-
ings. That is, CSA equals the change in retained earnings minus net income and plus net
dividends. There are several problems with this approach, however. The main problem is
that stock dividends, share retirements and share repurchases (when accounted for using
the par value method) reduce retained earnings but are not CSA.
22
Comprehensive Net Income (CNI) = net income (#172) minus preferred dividends (#19)
and plus Clean Surplus Adjustment to net Income (CSA).
Comprehensive Operating Income (OI) = Comprehensive Net Financial Expense (NFE)
plus Comprehensive Net Income (CNI) and plus Minority Interest in Income (MII, #49).
Some parts of dirty surplus income (e.g., currency translation gains and losses) are associ-
ated with minority interests, but COMPUSTAT (and often nancial reports) do not provide
the details necessary to adjust minority interest income for dirty surplus items. So minority
interest income, and therefore operating income, are misstated. However, since minority
interest is usually very small, the effect of this misstatement is likely to be immaterial for
most companies.
Unusual Operating Income (UOI) = after tax nonoperating income (expense) excluding
interest and equity in earnings ((#190 #55) (1 marginal tax rate)) plus after
tax special items (#17 (1 marginal tax rate)) plus extraordinary items & discon-
tinued operations (#48) plus cumulative translation adjustment (#230) and minus lag
cumulative translation adjustment (lag #230).
Other Operating Income Items (Other Items) = Equity in earnings (#55).
Operating Income from Sales (OI from Sales) = Operating Income (OI) minus Other
Operating Income Items (Other Items).
Core Operating Income from Sales (Core OI from Sales) = Operating Income from Sales
(OI from Sales) minus Unusual Operating Income (UOI).
Marginal Tax Rate =the top statutory federal tax rate plus 2%average state tax rate. The top
federal statutory corporate tax rate was 52% in 1963, 50% in 1964, 48% in 19651967,
52.8% in 19681969, 49.2% in 1970, 48% in 19711978, 46% in 19791986, 40% in
1987, 34% in 19881992 and 35% in 19931999.
Interest on Operating Obligations (io) = the one year risk-free rate at the beginning of
the year multiplied by the difference between operating liabilities (OL) and Deferred
Taxes and Investment Tax Credit (#35).
Amore precise calculation of io would not impute implicit interest on the accrued pension
cost but instead add the net interest cost on the net pension obligation (i.e., interest cost
minus actual return on plan assets). However, Compustat gives details about the pension
interest cost and the actual return on plan assets only starting 1991, and it does not provide
the net accrued pension cost. This liability is probably included in item #75 (liabilities
other).
Acknowledgments
The comments of Trevor Harris, Charles Lee, JimOhlson (and his teaching notes), Katherine
Schipper and Toshi Shibano were helpful.
Notes
1. Variation in outcomes for different scenarios also indicates risk.
2. For nancial institutions, nancial assets and liabilities are operating assets and liabilities. Financial insti-
tutions are excluded from our empirical analysis.
3. Johannson (1998) distinguishes operating andnancing liabilities and the role they play inprotability analysis,
but does not develop the formal analysis of operating liability leverage here.
4. Deferred taxes are not discounted (and the government does not implicitly charge interest). Net interest cost
of pension liabilities is explicit in the income statement. So, with these items on the balance sheet,
io = short-term borrowing rate OL other than deferred taxes and pensions
+net interest cost on pension liabilities.
5. When there are minority interests, an additional driver is the ratio of minority interests to common equity.
6. For V
NOA
0
and V
NFO
0
to be added as separate values and be equivalent to (1), it is required that
E
=
W
+
V
NFO
0
V
E
0
[
W

D
]
(which is the same form as (4)). That is, the cost of capital for operations is
W
=
V
E
0
V
NOA
0
V
NFO
0
V
NOA
0
as in standard nance that evokes Modigliani and Miller nancing irrelevance (without consideration of tax
effects fromthe nancing). The distinction between different costs of capital for different net assets is somewhat
cosmetic for the purposes of this paper, as the ratio analysis and the documentation that follows do not concern
the cost of capital. The observation that different income streams have different risks (and thus different costs
of capital) is standard in nance texts and we adhere to this orthodoxy largely so as to not offend the eye.
7. Growth in residual earnings translates into a P/E ratio, the multiple commonly used to indicate growth.
Forecasting constant RE yields a normal P/E ratio (equal to
E
/(
E
1)) and forecasting growing RE yields
a P/E ratio greater than normal. See Penman (1996).
8. Free cash ow treats growth in NOA (NOA) as a negative driver (it is subtracted in (14)) whereas residual
income treats investment as a positive driver. This identication is the main difference between residual income
and discounted cash ow techniques.
9. Again, if minority interests exits, equation (16) is a decomposition of ROTCE, not ROCE.
10. The analysis omits effects on shareholder value of taxes on dividends and the value effects of the tax-
deductibility of interest on debt, if any. GAAP omits the calculation of the cost to shareholders of employee
stock options but this is a matter of recognizing an expense in comprehensive income. It does not affect the
decomposition.
11. Except for the calculation of NBC and Core NBC, since rms can have negative net nancial obligations (that
is, net nancial asset positions).
12. Some rms adopted the Statement in 1993.
13. The median growth rates for RE and ReOI in Table 1 are negative. Growth rates are calculated only for the
cases of positive denominators. It will become clear in the time-series analysis that follows that these residual
income measures subsequently decline, on average, for rms with positive residual income.
14. The product-moment (Pearson) correlation between median ROCE and the treasury yield is 0.32. The correla-
tion between the lagged median ROCE and the treasury yield is 0.67. The one-year treasury yield is measured
at the beginning of the year. Thus ROCE is more strongly related to the yield at the end of the year than to the
yield at the beginning of the year.
15. In constructing these gures, a few extreme observations were deleted.
16. Ohlson and Zhang (1999) and Penman (1997) model nite-horizon residual earnings valuation.
17. See, for example, Frankel and Lee (1998), Lee, Myers and Swaminathan (1999), Penman and Sougiannis
(1998), and Francis, Olsson and Oswald (2000).
18. Survivorship bias is a particular concern in interpreting Figures 4c and 4d, however. CSE and NOA increase
through acquisitions and such growth is incorporated in the numbers here. But the negative growth of rms
acquired is not.
19. Figure 5b includes only non-zero Other Core Items.
20. The pattern for the largest negative change in Core Sales PM in year 0 is particularly interesting. This portfolio
has the largest (positive) changes in subsequent years. Is this big bath accounting that shifts income to the
future?
21. The persistence of ATO changes for one period here is partly due to using average NOA in the calculation of
ATO.
22. The effect of stock dividends on retained earnings is included in item #231 (Retained EarningsOther
Adjustments). This item also includes the effect of notes receivable (presumably for share issues). Both
items are not income items, and thus the change in #231 should be the deducted from the change in retained
earnings. The problem is that this item is available only starting 1982.
References
Abarbanell, J. S., and V. L. Bernard. (2000). Is the U.S. Stock Market Myopic? Journal of Accounting Research
38, 221243.
Beaver, W. H., and S. G. Ryan. (2000). Biases and Lags in Book Value and their Effects on the Abil-
ity of the Book-to-Market Ratio to Predict Book Return on Equity. Journal of Accounting Research 38,
127148.
Brief, R. P., and R. A. Lawson. (1992). The Role of the Accounting Rate of Return in Financial Statement
Analysis. The Accounting Review 67, 411426.
Brown, L. D. (1993). Earnings ForecastingResearch: Its Implications for Capital Markets Research. International
Journal of Forecasting 9, 295320.
Brown, S. J., W. N. Goetzmann, and S. A. Ross. (1995). Survival. Journal of Finance 50, 853873.
Claus, J. J., and J. K. Thomas. (2001). Equity Premium as Low as Three Percent? Evidence from Analysists
Earnings Forecasts for Domestic and International Stocks Journal of Finance. Forthcoming.
Colander, D. (1992). Retrospectives: The Lost Art of Economics. Journal of Economic Perspectives 6, 191198.
Edwards, E. O., and P. W. Bell. (1961). The Theory and Measurement of Business Income. Berkeley, CA: The
University of California Press.
Faireld, P. M., R. J. Sweeney, and T. L. Yohn. (1996). Accounting Classication and the Predictive Content of
Earnings. The Accounting Review 71, 337355.
Fama E. F., and K. R. French. (2000). Forecasting Protability and Earnings. Journal of Business 73.
Feltham, G. A., and J. A. Ohlson. (1995). Valuation and Clean Surplus Accounting for Operating and Financial
Activities. Contemporary Accounting Research 11, 689731.
Francis, J., P. Olsson, and D. R. Oswald. (2000). Comparing the Accuracy and Explainability of Dividend, Free
Cash Flow, and Abnormal Earnings Equity Value Estimates. Journal of Accounting Research 38, 4570.
Frankel, R., and C. M. C. Lee. (1998). Accounting Valuation, Market Expectation and Cross-Sectional Stock
Returns. Journal of Accounting and Economics 25, 283319.
Freeman, R. N., J. A. Ohlson, and S. H. Penman. (1982). Book Rate-of-Return and Prediction of Earnings
Changes: An Empirical Investigation. Journal of Accounting Research 20, 639653.
Gebhardt, W. R., C. M. C. Lee, and B. Swaminathan. (2000). Toward an Implied Cost of Capital. Journal of
Accounting Research, Forthcoming.
Graham, B., D. L. Dodd, and S. Cottle. (1962). Security Analysis. New York, NY: McGraw-Hill Book Company,
Inc.
Ibbotson, R., and R. Sinqueeld. (1983). Stocks, bonds, bills and ination. 19261982, Charlottesville, VA:
Financial Analyst Research Foundation.
Johansson, S. (1998). The Protability, Financing, and Growth of the Firm: Goals, Relationships and Measurement
Methods. Lund, Sweden: Studentlitteratur.
Kay, J. A. (1976). Accountants, too, Could be Happy in a Golden Age: The Accountants Rate of Prot and the
Internal Rate of Return. Oxford Economic Papers 17, 447460.
Lee C. M. C., J. Myers, and B. Swaminathan. (1999). What Is the Intrinsic Value of the Dow? Journal of Finance
54, 16931741.
Lev, B., and S. R. Thiagarajan. (1993). Fundamental Information Analysis. Journal of Accounting Research 31,
190215.
Lipe, R. C. (1986). The Information Contained in the Components of Earnings. Journal of Accounting Research
24, 3764.
Myers, J. N. (1999). Conservative Accounting and Finite Firm Life: Why Residual Income Valuation Estimates
Understate Stock Price. Working Paper, University of Washington.
Ohlson, J. A. (1995). Earnings, Book Values, and Dividends in Equity Valuation. Contemporary Accounting
Research 11, 661687.
Ohlson J. A., and X. J. Zhang. (1999). On the Theory of the Forecast-Horizon in Equity Valuation. Journal of
Accounting Research 37, 437449.
OHanlon J., and A. Steele. (1998). Estimating the Equity Risk Premium Using Accounting Fundamentals.
Working Paper, Lancaster University.
Ou, J. A. (1990). The Information Content of Nonearnings Accounting Numbers as Earnings Predictors. Journal
of Accounting Research 28, 144163.
Ou, J. A., and S. H. Penman. (1989). Financial Statement Analysis and the Prediction of Stock Returns. Journal
of Accounting and Economics 11, 295329.
Ou, J. A., and S. H. Penman. (1995). Financial Statement Analysis and the Evaluation of Market-to-Book Ratios.
Working Paper, University of California at Berkeley.
Penman, S. H. (1996). The Articulation of Price-Earnings Ratios and Market-to-Book Ratios and the Evaluation
of Growth. Journal of Accounting Research 34, 235259.
Penman, S. H. (1991). An Evaluation of Accounting Rate-of-Return. Journal of Accounting, Auditing, and
Finance 6, 233255.
Penman S. H. (1997). A Synthesis of Equity Valuation Techniques and the Terminal Value Calculation of the
Dividend Discount Model. Review of Accounting Studies 2, 303323.
Penman S. H., and T. Sougiannis. (1998). A Comparison of Dividend, Cash Flow, and Earnings Approaches to
Equity Valuation. Contemporary Accounting Research 15, 343383.
Preinreich, G. (1938). Annual Survey of Economic Theory: The Theory of Depreciation. Econometrica, 219
241.
Selling, T. I., and C. P. Stickney. (1989). The Effects of Business Environments and Strategy on a Firms Rate of
Return on Assets. Financial Analysts Journal, 4352.
Zhang, X. (2000). Conservative Accounting and Earnings Valuation. Journal of Accounting and
Economics 29, 125149.

Nissim and Penman - Ratio Analysis and Equity Valuation From Research To Practice

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nissim and Penman - Ratio Analysis and Equity Valuation From Research To Practice

Uploaded by

Copyright:

Available Formats

Review of Accounting Studies, 6, 109154, 2001

C 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.

= ROOA +[OLLEV OLSPREAD]

Sales Prot Margin (Sales PM)

Asset Turnover (ATO)

Financial Leverage (FLEV)

Operating Liability Leverage (OLLEV)

Minority Interest Sharing (MSR)

Asset Turnover (ATO)

Financial Leverage (FLEV).

Sales Prot Margin (Sales PM)

Asset Turnover (ATO)

Operating Liability Leverage (OLLEV)

Asset Turnover (ATO)

Core RNOA Core NBC +

= (Core OI from Sales + io)/Sales, ATO

which excludes operating liabilities.

Ratio analysis is integrated with valuation analysis, giving substance to fundamental

The analysis is of comprehensive income. Components of other comprehensive income

There is an extended analysis of drivers of protability beyond the standard analysis.

Minority interest share in accounting value is accommodated.

Financial leverage is redened from the traditional Debt/Equity ratio.

You might also like